Subjects studied a mixed list of 70 words that varied in imagery value and then received three successive tests. Also varied were instructions given to subjects prior to list presentation (imagery or semantic rehearsal) and the type of recall test (standard free recall, an uninhibited-recall procedure in which subjects were encouraged to free associate and to guess while recalling the list, and a forced-recall condition in which they were also told to write a large number of responses to fill the allotted spaces). Recall improved across the three tests in all conditions, but the improvement was greater for high-than for low-imagery words. In addition, hypermnesia (the improved recall across tests) was shown to occur following semantic rehearsal instructions as well as imagery instructions and to occur with low-imagery words, contrary to the imagery hypothesis of the effect. Most importantly, the large variation in recall criterion produced by manipulating instructions at test (as measured by intrusions) did not affect the overall level of correct recall or the magnitude of improvements across tests. Apparently, the assumption of generate-recognize theories that people generate much more information in free recall than they produce (due to a stringent criterion for recognition of the generated material) is false.Hypermnesia refers to the improvement in recall across repeated tests. The term was introduced into the modern literature by Erdelyi and Becker (1974) as a replacement for the term reminiscence (used by Ballard, 1913, in describing his experiments), because reminiscence had come to have multiple meanings. Some researchers used reminiscence to refer to absolute improvements across tests, whereas others applied it to recall of events that were forgotten on one test but recovered on a later one. (This latter sense was the one originally used by Ballard.) In this paper we retain the usage, suggested by Roediger and Thorpe (1978), of reminiscence as recovery of previously forgotten items on a later test (intertest recovery) and hypermnesia as the absolute increase of recall across tests (when intertest recovery exceeds intertest forgetting).The finding of improvements in recall across repeated tests has suffered several reversals of fortune in this century, with some researchers finding it and others not (see Woodworth, 1929, pp. 63-68, and McGeoch & Irion, 1952, pp. 140-148, for the early history). However, in recent years the procedure introduced by Erdelyi and his colleagues has provided a reliable method for demonstrating and studying the phenomenon (e.g., Erdelyi & Becker, 1974;Erdelyi, Buschke, & Finkelstein, 1977;Erdelyi, Finkelstein, Herrell, Miller, & Thomas, 1976 successive recall tests of 7-min duration each. Some of the main findings to emerge from Erdelyi's research cited above are that: (1) recall improves for pictures but not for concrete words across the three tests; (2) study of concrete words following instructions to image the words produces hypermnesia; and (3) generating words at study fr...